Cosmological simulations of quasar fueling to sub-parsec scales using Lagrangian hyper-refinement

Page created by Jordan Bishop
 
CONTINUE READING
Draft version August 31, 2020
                                                 Typeset using LATEX twocolumn style in AASTeX63

                                                                     Cosmological simulations of quasar fueling to sub-parsec scales
                                                                                  using Lagrangian hyper-refinement
                                                                Daniel Anglés-Alcázar,1, 2 Eliot Quataert,3, 4 Philip F. Hopkins,5 Rachel S. Somerville,2, 6 Christopher C. Hayward,2
                                                                   Claude-André Faucher-Giguère,7 Greg L. Bryan,8, 2 Dušan Kereš,9 Lars Hernquist,10 and James M. Stone11
                                                                        1 Department    of Physics, University of Connecticut, 196 Auditorium Road, U-3046, Storrs, CT 06269-3046, USA
                                                                           2 Center   for Computational Astrophysics, Flatiron Institute, 162 Fifth Avenue, New York, NY 10010, USA
arXiv:2008.12303v1 [astro-ph.GA] 27 Aug 2020

                                                                  3 Department of Astronomy and Theoretical Astrophysics Center, University of California Berkeley, Berkeley, CA 94720, USA
                                                                                      4 Department of Astrophysical Sciences, Princeton University, Princeton, NJ 08544, USA
                                                                                      5 TAPIR, Mailcode 350-17, California Institute of Technology, Pasadena, CA 91125, USA
                                                                          6 Department of Physics and Astronomy, Rutgers University, 136 Frelinghuysen Rd, Piscataway, NJ 08854, USA
                                                                  7 CIERA and Department of Physics and Astronomy, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA
                                                                               8 Department of Astronomy, Columbia University, 550 West 120th Street, New York, NY 10027, USA
                                                                       9 Department of Physics, CASS, University of California at San Diego, 9500 Gilman Drive, La Jolla, CA 92093, USA
                                                                                  10 Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, MA 02138, USA
                                                                                 11 School of Natural Sciences, Institute for Advanced Study, 1 Einstein Drive, Princeton, NJ 08540

                                                                                                              Abstract
                                                            We present cosmological hydrodynamic simulations of a quasar-mass halo (Mhalo ≈ 1012.5 M at z = 2) that
                                                         for the first time resolve gas transport down to the inner 0.1 pc surrounding the central massive black hole. We
                                                         model a multi-phase interstellar medium including stellar feedback by supernovae, stellar winds, and radiation,
                                                         and a hyper-Lagrangian refinement technique increasing the resolution dynamically approaching the black hole.
                                                         We do not include black hole feedback. We show that the sub-pc inflow rate (1) can reach ∼6 M yr−1 roughly
                                                         in steady state during the epoch of peak nuclear gas density (z ∼ 2), sufficient to power a luminous quasar, (2) is
                                                         highly time variable in the pre-quasar phase, spanning 0.001–10 M yr−1 on Myr timescales, and (3) is limited
                                                         to short (∼2 Myr) active phases (0.01–0.1 M yr−1 ) followed by longer periods of inactivity at lower nuclear gas
                                                         density and late times (z ∼ 1), owing to the formation of a hot central cavity. Inflowing gas is primarily cool,
                                                         rotational support dominates over turbulence and thermal pressure, and star formation can consume as much gas
                                                         as provided by inflows across 1 pc–10 kpc. Gravitational torques from multi-scale stellar non-axisymmetries
                                                         dominate angular momentum transport over gas self-torquing and pressure gradients, with accretion weakly
                                                         dependent on black hole mass. Sub-pc inflow rates correlate with nuclear (but decouple from global) star
                                                         formation and can exceed the Eddington rate by ×10. The black hole can move ∼10 pc from the galaxy center
                                                         on ∼0.1 Myr. Accreting gas forms pc-scale, rotationally supported, obscuring structures often misaligned with
                                                         the galaxy-scale disk. These simulations open a new avenue to investigate black hole–galaxy co-evolution.

                                                                          1. Introduction                                          2013; Cicone et al. 2014; Harrison et al. 2014; Zakamska &
                                                  The inflow of gas from large scales down to galactic nuclei                      Greene 2014; Ramos Almeida et al. 2017) and radio-emitting
                                               plays a key role in galaxy formation, driving the growth of central                 jets (Fabian 2012; Hlavacek-Larrondo et al. 2012) that may have
                                               massive black holes and a variety of related phenomena: from                        a significant impact on galaxy evolution (e.g. Silk & Rees 1998;
                                               bright quasars (QSOs) that outshine their host galaxies (with                       Di Matteo et al. 2005; Murray et al. 2005; Faucher-Giguère &
                                               bolometric luminosities reaching Lbol ∼ 1046 –1048 erg s−1 , e.g.                   Quataert 2012; Richings & Faucher-Giguère 2018). The scaling
                                               Fan et al. 2001; Mortlock et al. 2011; Trakhtenbrot et al. 2011;                    relations between central black hole mass and properties of their
                                               Bañados et al. 2018; Zakamska et al. 2019) to active galactic                       host galaxies (e.g. Häring & Rix 2004; Hopkins et al. 2007;
                                               nuclei (AGN) “feedback” in the form of fast nuclear outflows                        Bentz et al. 2009; Bennert et al. 2011; Kormendy & Ho 2013;
                                               (e.g. Tombesi et al. 2013; Nardini et al. 2015), galaxy-scale winds                 McConnell & Ma 2013; Reines & Volonteri 2015; Graham 2016)
                                               (e.g. Rupke & Veilleux 2011; Greene et al. 2012; Liu et al.                         and the similarity between the global cosmic histories of star
                                                                                                                                   formation and black hole accretion (Silverman et al. 2008; Aird
                                                                                                                                   et al. 2010; Rodighiero et al. 2010; Heckman & Best 2014;
                                               Corresponding author: Daniel Anglés-Alcázar                                         Madau & Dickinson 2014) further suggest some form of black
                                               angles-alcazar@uconn.edu                                                            hole–galaxy co-evolution over cosmological timescales. It is
                                                                                                                                   thus crucial to understand the mechanisms driving gas inflows
2                                                    Daniel Anglés-Alcázar et al.

from galactic scales down to the black hole accretion disk in a       Prieto et al. 2017; McAlpine et al. 2018; Trebitsch et al. 2018;
full cosmological context, which remains a major challenge.           Lupi et al. 2019) but the details crucially depend on resolution,
   Modern large volume cosmological hydrodynamic simulations          ISM physics, and black hole parameterization.
such as Magneticum (Hirschmann et al. 2014), Horizon-AGN                 Idealized (non-cosmological) simulations offer the possibil-
(Dubois et al. 2014; Volonteri et al. 2016), Eagle (Schaye et al.     ity to study gas transport in galaxies in a more controlled setup
2015; Rosas-Guevara et al. 2016), Illustris (Genel et al. 2014;       and at much higher resolution than typically available in cos-
Vogelsberger et al. 2014a,b; Sijacki et al. 2015), BlueTides          mological simulations. Early models showed that large-scale
(Feng et al. 2016), Romulus (Tremmel et al. 2017; Sharma et al.       tidal torques induced by galaxy mergers or bar/spiral wave insta-
2020), IllustrisTNG (Weinberger et al. 2017; Pillepich et al. 2018;   bilities in self-gravitating disks can lead to angular momentum
Habouzit et al. 2019), and SIMBA (Davé et al. 2019; Thomas            transport and rapid inflow of gas to the central sub-kpc of galaxies
et al. 2019; Borrow et al. 2020) have implemented sub-grid mod-       (Hernquist 1989; Barnes & Hernquist 1991; Hernquist & Mihos
els of black hole growth and feedback with increasing success         1995; Barnes & Hernquist 1996), confirmed by more recent sim-
at reproducing global galaxy properties and black hole observ-        ulations (e.g. Springel et al. 2005; Younger et al. 2008; Hopkins
ables. However, despite much recent progress, the typical ∼kpc        et al. 2009). Gas inflow to
Cosmological hyper-refinement simulations of quasar fueling                                                      3

stellar components, gas consumption by star formation, turbu-          consortium2, which attempts to reduce the uncertainties associ-
lence and gas ejection driven by stellar feedback, and non trivial     ated with sub-grid parameterizations of key baryonic processes
black hole dynamics in the nuclear potential. We achieve this by       in galaxy formation to improve the accuracy of fundamental cos-
implementing multi-phase ISM physics from the FIRE project, a          mological measurements. Other work in SMAUG has centered
novel hyper-Lagrangian refinement technique that increases the         on star formation and stellar feedback (Kim et al. 2020; Li et al.
resolution dynamically closer to the black hole, and a sub-pc          2019, 2020; Motwani et al. 2020), the physics of the CGM (Field-
scale treatment of black hole growth that avoids many of the           ing et al. 2020), and the connection between semi-analytic mod-
uncertainties introduced by parameterized sub-grid models.             els and cosmological hydrodynamic simulations (Pandya et al.
   As a first application of this methodology, we focus on un-         2020). Our work complements these by providing a key step to-
derstanding the mechanisms responsible for gas transport across        wards developing better sub-grid models for black hole accretion
spatial scales and black hole growth during qualitatively distinct     in galaxy formation, which play a key role in the evolution of
phases of a massive galaxy (M? ∼ 6–20 ×1010 M ) before, dur-           massive galaxies and large-scale structure.
ing, and after its peak of nuclear gas density at z ∼ 2.5 → 1.
                                                                                                     2. Methods
This allows us to investigate the physical conditions conducive
to a broad range of black hole accretion properties, from large                   2.1. Cosmological galaxy formation model
QSO-like inflow rates down to extended periods of inactivity,             We use the FIRE-2 galaxy formation model presented in Hop-
which represents a unique challenge for current models. Given          kins et al. (2018a), which is an updated numerical implemen-
the complex multi-physics involved, we choose to neglect the           tation of the original FIRE simulations (Hopkins et al. 2014a)
effects of black hole feedback in an attempt to reduce model           with more accurate hydrodynamics, gravitational softening, and
uncertainties and avoid making prior assumptions about the effi-       supernova coupling algorithms. FIRE-2 employs the GIZMO3
ciency of AGN feedback. The results presented here provide the         N-body+hydrodynamics code (Hopkins 2015) in its “meshless
baseline for future simulations including AGN feedback.                finite mass" (MFM) mode. MFM is a mesh-free, Lagrangian
   We describe our methodology in §2, focusing on the galaxy           finite-volume Godunov method that combines the advantages of
formation model (§2.1) and the use of pre-existing FIRE cos-           particle-based and grid-based methods: MFM provides auto-
mological zoom-in simulations as the initial conditions (§2.2)         matic adaptivity in resolution while minimizing advection and
to implement our new hyper-Lagrangian refinement technique             angular momentum conservation errors relative to AMR codes
(§2.3). We present an overview of our simulations in §3, il-           and exhibits superior performance in shock-capturing and fluid-
lustrating the radical increase in resolution relative to previous     mixing problems while avoiding the low-order errors inherent
models (§3.1) and their unique dynamic range (§3.2). We then           to smooth particle hydrodynamics (SPH). As shown in Hop-
quantify the galaxy radial structure and thermodynamic state of        kins (2015), MFM performs well with different particle masses
gas over five orders of magnitude in spatial scales (§3.3), which      unlike many SPH methods where the errors increase when hav-
provides significant insight into the physics driving black hole       ing unequal-mass particles interacting, which is crucial for our
growth and the connection between accretion and galaxy prop-           hyper-refinement simulations.
erties on larger scales. Our main results are presented in §4,            We adopt a “standard" flat ΛCDM cosmology with parameters
where we measure for the first time resolved accretion rates at        H0 = 69.7 km s−1 Mpc−1 , ΩM = 1 − ΩΛ = 0.2821, Ωb = 0.0461,
0.1 pc under different conditions at the center of a massive galaxy,   σ8 = 0.817, and ns = 0.9646 (Hinshaw et al. 2013). Gravita-
reaching peak inflow rates sufficient to power a luminous quasar       tional forces are computed using an improved version of the tree-
(§4.1). We study the connection between star formation and gas         particle-mesh gravity solver of the GADGET-3 code (Springel
inflow rate across the full range of spatial scales 0.1 pc–10 kpc      2005), including adaptive gravitational softenings for the gas
(§4.2), with important implications for black hole fueling and         component. Following Price & Monaghan (2007), we assume
the star formation-AGN connection. We quantify the angular             that the mass distribution corresponding to each element has the
momentum properties of gas across scales (§4.3) and evaluate           same functional form as the interaction kernel in MFM. The
the relative contributions of gravitational torques and pressure       time integration is fully adaptive with a power-of-two hierarchy
gradients to angular momentum transport (§4.4), a crucial step         for assigning individual time-steps (Springel 2005), including
toward identifying the dominant mechanisms driving gas inflows.        a time-step limiter to handle strong feedback events (Durier &
We further study the intrinsic dependence of accretion rate on         Dalla Vecchia 2012).
black hole mass (§4.5), with important implications for sub-grid
black hole accretion prescriptions. We discuss our results and                  2.1.1. Cooling, star formation, and stellar feedback
conclude in §5. Additional details about the robustness of our           We incorporate radiative cooling and heating processes from
methodology are presented in Appendix A, including systematic          T = 10–1010 K as detailed in Hopkins et al. (2018a), includ-
tests of resolution convergence (A.1) and other numerics (A.2).        ing free-free, photoionization/recombination, Compton, photo-
   This work has been developed with the support of the Simulat-       electric, metal-line, molecular, fine-structure, dust collisional,
ing Multiscale Astrophysics to Understand Galaxies (SMAUG)

                                                                         2 https://www.simonsfoundation.org/flatiron/center-for-computational-astrophysics/
                                                                           galaxy-formation/smaug/
                                                                         3 http://www.tapir.caltech.edu/~phopkins/Site/GIZMO.html
4                                                                 Daniel Anglés-Alcázar et al.

and cosmic ray4 processes. Star formation occurs only in gas                             here. The same feedback implementation has been successfully
which is (1) locally self-gravitating following the sink-particle                        used in cosmological zoom-in simulations with mass resolution
criterion of Hopkins et al. (2013), (2) molecular with self-                             ranging from mb ∼ 20 M in dwarfs (Wheeler et al. 2019) to
shielding fraction given by the local Sobolev approximation                              mb ∼ 3 × 104 M in massive galaxies (Anglés-Alcázar et al.
(Krumholz & Gnedin 2011), (3) Jeans unstable with gas ele-                               2017c). Stellar winds follow a similar implementation but with
ment mass larger than the thermal Jeans mass, and (4) above                              continuous rather than impulsive injection of mass, momentum,
a minimum hydrogen number density nH ≥ 1000 cm−3 . Eligi-                                energy, and metals into surrounding gas particles. The discrete
ble gas elements are converted into collisionless star particles                         nature of individual SNe is crucial in generating a multi-phase
with a probability set by their SFR density ρÛ? integrated over a                        ISM and driving galactic winds, but IMF sampling on radia-
timestep, where ρÛ? = ρmol /tff and we assume 100% efficiency                            tive feedback and stellar winds has weak effects on galaxy-scale
per local free-fall time tff . The global efficiency of star forma-                      properties (Su et al. 2018).
tion is self-regulated by stellar feedback and is insensitive to the
details of the star formation model (e.g., Faucher-Giguère et al.                                             2.1.2. Black hole physics
2013; Hopkins et al. 2014a; Orr et al. 2018).                                               We follow the growth of a massive black hole located at
   Each star particle represents a single stellar population with                        the center of the main halo in the high resolution region of
known mass, age, and metallicity. Stellar feedback is imple-                             a pre-existing cosmological zoom-in simulation from Anglés-
mented at the scale of star-forming regions following explicitly                         Alcázar et al. (2017c), which implemented a sub-grid accretion
the time dependence of several different mechanisms, including                           model based on gravitational torques (Hopkins & Quataert 2011;
energy, momentum, mass, and metal injection from (1) Type Ia                             Anglés-Alcázar et al. 2017a) evaluated on ∼ 100 pc scales. The
and Type II Supernovae (SNe), (2) continuous stellar mass-loss                           radical increase in resolution achieved by the hyper-Lagrangian
from OB and AGB winds, (3) photoionization and photoelectric                             refinement technique introduced here allows us to study black
heating, and (4) local and long-range momentum flux from radia-                          hole growth without the need for a sub-grid parameterization for
tion pressure. All feedback quantities and their time dependence                         the inflow rate from larger scales down to the accretion disk.
are tabulated directly from the stellar population synthesis model                          The black hole is represented by a collisionless particle with
starburst99 (Leitherer et al. 1999) assuming a Kroupa (2001)                             initial mass MBH = 108 M , which is three orders of magnitude
initial mass function (IMF).                                                             larger than the mass of dark matter particles and many orders of
   The radiative feedback implementation is described in detail                          magnitude larger than the high resolution gas and star particles
in Hopkins et al. (2020a). The age and metallicity-dependent                             that dominate the potential in the central ∼100 pc. This allows
IMF-averaged spectrum of each star particle is locally extin-                            us to track the black hole dynamics explicitly owing to resolved
guished by the surrounding gas using a Sobolev approxima-                                gravitational forces without the need for a sub-grid dynamic fric-
tion with frequency and metallicity-dependent opacities. The                             tion prescription (e.g. Tremmel et al. 2015) or the artificial drag
emerging luminosities (including re-radiated dust emission) are                          forces and re-positioning algorithms required in large volume
propagated through a tree structure in the optically thin limit,                         cosmological simulations (Di Matteo et al. 2008; Dubois et al.
yielding a long-range incident flux which is then corrected for                          2014; Beckmann et al. 2018; Davé et al. 2019).
local extinction at the location of the gas element. We also in-                            We model accretion by direct gravitational capture of gas by
clude a uniform, redshift-dependent photo-ionizing background                            the central black hole. Gas particles located within the accretion
(Faucher-Giguère et al. 2009) with self-shielding. The result-                           radius Racc are captured if (1) their velocity relative to the black
ing incident flux is used to compute the ionization state of the                         hole is lower than the escape velocity and (2) the apocentric
gas, radiative heating/cooling rates, and radiation pressure from                        radius of the particle relative to the black hole is also within
photon absorption including UV/optical single-scattering and re-                         Racc . We choose the accretion radius Racc ≡ 0.1 pc so that it
radiated infrared photons.                                                               roughly corresponds to the scale below which the accretion disk
   The SNe feedback algorithm, fully described in Hopkins et al.                         is expected to form, Racc ∼ 104 Rs (e.g. Goodman 2003), where
(2018b), is constructed to ensure conservation of mass, energy,                          Rs = 2 G MBH /c2 ≈ 10−5 pc is the Schwarzschild radius for
and momentum as well as isotropic injection in the rest frame                            MBH = 108 M . We do not allow the black hole to accrete star
of the star particle regardless of the geometry and dynamics                             particles. A similar explicit gravitational capture approach was
of the surrounding gas distribution (non-trivial in Lagrangian                           used in Hopkins & Quataert (2011, 2010) and Hopkins et al.
codes). Each SN explosion is treated independently, injecting                            (2016) to study black hole growth in idealized nuclear scale
the correct amount of momentum and thermal energy depend-                                simulations. For simplicity, we refer to the measured gas inflow
ing on the resolved coupling distance relative to the SN cool-                           rate through Racc as “black hole accretion rate" or “ MÛ BH ” but
ing radius. This algorithm reproduces converged solutions in                             it should be interpreted as the instantaneous feeding rate onto
both energy and momentum independent of resolution, which                                the black hole accretion disk. This represents an upper limit
is crucial for the multi-mass resolution simulations presented                           to the actual black hole growth owing to e.g. mass loss in
                                                                                         accretion disk winds (e.g. Proga et al. 2008; Yuan et al. 2012;
                                                                                         Jiang et al. 2014). We neglect the effects of black hole feedback
    4 We include approximate cosmic ray heating in the dense ISM but do not include
      full cosmic ray transport and feedback as in e.g. Chan et al. (2019) and Hopkins
                                                                                         in an attempt to understand the mechanisms responsible for mass
      et al. (2020b).                                                                    transport without making any assumptions about the efficiency
                                                                                         of AGN feedback.
Cosmological hyper-refinement simulations of quasar fueling                                                                 5

         2.2. Zoom-in simulations as initial conditions                                                                    z
                                                                                                10 7 6 5   4   3               2                          1
   We use the FIRE-2 zoom-in simulations of massive halos
(Mhalo ∼ 1012.5−13 M at z=1) presented in Anglés-Alcázar                                   11

                                                                        log10(Mstar/M )
et al. (2017c) to identify interesting redshift snapshots for re-
simulation at ultra-high resolution. The initial conditions corre-                         10
spond to a subset of the A series of MassiveFIRE halos that were
originally simulated with the FIRE-1 model (Feldmann et al.                                9

                                                                                                                                               late-AGN
                                                                                                                   pre-QSO
                                                                                                                   full-QSO
2016, 2017). For this work, we focus on halo A4 (studied also
in Cochrane et al. 2019 and Wellons et al. 2020), which was                                8
evolved from early times down to z = 1 including an on-the-
fly treatment of black hole growth but not black hole feedback.
                                                                                           3

                                                                       log10(SFR/M yr 1)
The mass resolution employed was mb = 3.3 × 104 M and                                      2
mDM = 1.7 × 105 M for the baryonic and dark matter compo-
nents. The force softenings were gas min = 0.7 pc,  = 
                                                     ?    BH = 7 pc,                       1
and DM = 57 pc, where gas min is the minimum adaptive force soft-

ening for gas (with gas identical to the kernel smoothing scale)                          0
and the softenings for the stellar (?), black hole (BH ), and dark
matter (DM ) components are fixed in physical units at z < 9.                             1
   Figure 1 shows the global stellar mass (M?; top) and SFR                                11
                                                                                              R < 100pc

                                                                       kpc 2)
(middle) as a function of redshift for the central galaxy of halo
A4. Throughout this paper, global galaxy properties refer to                               10
integrated quantities within 0.1 Rvir of the host halo. We iden-                            9
tify halos using the Amiga Halo Finder (AHF; Gill et al. 2004;         gas/M                8
Knollmann & Knebe 2009), modified such that the virial ra-
                                                                                            7
                                                                       log10(
dius Rvir corresponds to the evolving overdensity definition of
Bryan & Norman (1998). We also show the nuclear gas sur-                                    6
face density (Σgas ; bottom) measured within the black hole ker-                                 1      2              3           4       5              6
nel R0 ∼ 100 pc. While R0 varies with time as the galaxy                                                       Cosmictime (Gyr)
evolves (formally defined to contain 256 gas resolution ele-
ments), Σgas (< R0 ) is representative of the typical nuclear gas        Figure 1. Stellar mass (top), SFR (middle), and nuclear gas surface
density within the inner ∼ 100 pc. Global galaxy properties are          density (within R0 ∼ 100 pc; bottom) as a function of redshift for
computed from the snapshots available at ∼ 20–25 Myr inter-              the central galaxy in halo A4 from Anglés-Alcázar et al. (2017c), with
vals, but we take advantage of black hole-specific data outputs          Mvir ∼ 5×1012 M at z = 1. Vertical lines indicate the redshifts chosen
that recorded the physical conditions within R0 (including Σgas )        for re-simulation with dynamic hyper-refinement of the gas component.
for every timestep of the simulation, reaching time resolution           We select: (1) z = 2.25 (orange) as the most optimistic conditions to
dt ∼ 100 yr.                                                             trigger a luminous “full-QSO” phase, with the highest nuclear Σgas
   The evolution of the central galaxy in halo A4 is qualitatively       and near the peak of global SFR; (2) z = 2.28 (green) as the less
similar to that of halo A2 described in detail in Anglés-Alcázar         extreme conditions prevailing ∼40 Myr before the full-QSO phase,
et al. (2017c). During the first ∼ 3 Gyr, the stellar mass grows         representative of “pre-QSO” conditions; and (3) z = 1.10 (blue) as the
to M? ∼ 5 × 1010 M while undergoing strong bursts of star                typical conditions with lower SFR and Σgas prevalent at late times, more
formation reaching SF R ∼ 10–300 M yr−1 . Correlated SNe                 representative of lower luminosity “late-AGN” conditions.
drive large scale winds that can evacuate a large fraction of ISM
gas and temporarily shut down star formation, while recycling              We focus on three qualitatively distinct host galaxy conditions
of wind material provides fuel for subsequent bursts of star for-        to study black hole feeding at ultra-high resolution, indicated in
mation (Muratov et al. 2015; Anglés-Alcázar et al. 2017b). At            Figure 1 by vertical lines of different colors:
later times, star formation becomes less bursty and galactic winds
less efficient (Stern et al. 2020), with the galaxy growing more                           • We identify z = 2.25 (orange) as the time at which
steadily to M? & 2 × 1011 M down to z = 1. This transition                                   galaxy A4 experienced the highest nuclear gas surface
from bursty to steady star formation has a significant impact on                             density, with Σgas (< 100 pc) > 1011 M kpc−2 and global
the nuclear gas reservoir: Σgas fluctuates by orders of magnitude                            SFR ∼200 M yr−1 . These extreme conditions are the
at early times but there is a more steady nuclear gas reservoir                              most likely to trigger gas inflow rates sufficient to power
at late times. Anglés-Alcázar et al. (2017c) showed that this                                a luminous quasar, which we denote as “full-QSO” phase.
has important implications for black hole growth, which can be
significantly suppressed at early times relative to the host galaxy                        • We investigate the conditions leading to the
growth (see also Çatmabacak et al. 2020 and Byrne et al. in                                  full-QSO phase by going ∼40 Myr back in time to
prep.).                                                                                      z = 2.28 (green), with higher global star formation,
                                                                                             SFR ∼500 M yr−1 , but lower nuclear gas density,
6                                                            Daniel Anglés-Alcázar et al.

                          BH radius of influence    log(MBH/M ) = 8          the gravity tree (Garrison-Kimmel et al. 2017), which allows us
                          BH/gas softening                                   to define a mass refinement factor χref (R) as an analytic function
                    5     BH accretion radius
                                                                             of R. Gas resolution elements split into two new elements if
                                                                             mg ≥ χref (R) × 2 mb for their current mass mg and distance R,
                                                                             where mb is the baryonic mass resolution in the original zoom-in
                    4
    log10(mg /M )

                                                                             simulation. The mass of each new split element is 0.5 mg and
                                                                             they are located at a distance dr = min(0.25 hsml, 0.35 dngb ) from
                                                                             the position of the parent resolution element, in opposite sides
                    3                                                        along a random direction. Here, hsml is the smoothing length,
                                                                             dngb is the distance to the nearest neighbor, and the numerical
                                                                             factors are chosen to minimize perturbations to hydrodynamic
                    2                                                        quantities and avoid overlapping of fluid elements. The velocity
                                                                             of split elements is that of the parent gas element with a first-order
                                                                             correction to account for their new location and the existing ve-
                    1 6        5        4       3 2 1        0      1
                                                                             locity gradient. Split elements are allowed to merge onto higher
                                                                             mass elements when they are significantly under-massive rela-
                                            log10(R/kpc)                     tive to the resolution requirement, if mg < χref (R) × mb /200,
                                                                             which is infrequent for the simulations presented here. All fluid
Figure 2. Gas mass resolution as a function of radial distance from          quantities are recomputed immediately after particle splitting or
the black hole for separate pre-QSO simulations with increasing hyper-       merging. Our fiducial simulations only refine the gas component
refinement level, from χrefmin = 1/2 (yellow) to χ min = 1/2048 (purple).
                                                   ref                       but we have also implemented splitting/merging of collisionless
For each radial bin, shaded regions correspond to the 10%–90% per-           particles (see Appendix A.2 for details).
centiles of the distribution. Black dotted lines show the refinement con-       Figure 2 shows the mass of gas resolution elements as a func-
dition mref ∝ R2 for different target mass resolution near the black hole,   tion of R obtained for separate simulations using different levels
normalized to match the original mass resolution of the cosmological         of refinement for our fiducial χref (R) functional form:
zoom-in simulation at R = 1 kpc. The vertical gray solid line indi-
cates the softening length of the black hole particle (BH = 0.001 pc),                    χmin × (1 − R/Rref ) + (R/Rref )2
                                                                                          
                                                                                          
                                                                                                                                if R ≤ Rref
                                                                                 χref =        ref                                             (1)
                                                                                          
which is also the minimum (adaptive) force softening length for gas
elements, while the vertical gray dashed line indicates the maximum                       1
                                                                                                                               if R > Rref
                                                                                          
size of the black hole kernel (Racc = 0.1 pc). Gas elements at R > Racc
                                                                                      min is the desired refinement factor near the black hole,
                                                                             where χref
cannot be accreted by the black hole. The vertical orange line in-
                                                                             which we vary from χref min = 1/2 (yellow) to χ min = 1/2048 (pur-
dicates the black hole radius of influence for MBH = 108 M and                                                              ref
pre-QSO conditions. Dynamic splitting improves the mass resolution           ple), and Rref = 1 kpc is the distance at which super-Lagrangian
from mg = 3 × 104 M → 15 M in our highest resolution simulations.            refinement begins to operate. The original baryonic mass res-
                                                                             olution mb = 3.3 × 104 M is retained at R > 1 kpc, decreas-
                                                                             ing rapidly down to mg ≈ χref × mb closer to the black hole.
            Σgas (< 100 pc) ∼ 1010 M kpc−2 . We denote these as
                                                                             In our highest resolution simulations we reach mg ≈ 15 M ,
            the “pre-QSO" initial conditions.
                                                                             which allows us to follow the inflowing gas well within the
       • We compare the z > 2 conditions near the peak of star for-          black hole radius of influence and down to the accretion aper-
         mation with the late time conditions at z = 1.10 (blue) with        ture Racc = 0.1 pc. Despite the strong gradient in resolution at
         significantly lower activity, where SFR ∼10 M yr−1 and              R = 0.1–1 kpc, we do not see any significant artifacts owing to
         Σgas (< 100 pc) < 109 M kpc−2 . These conditions are pre-           dynamical splitting and merging of gas elements, except for a
         sumably representative of more common, low-luminosity               short initial transient that we discard in the analysis. We have
         AGN, which we denote as “late-AGN".                                 experimented extensively with increasing the mass resolution
                                                                             slowly as the simulation evolves and also refining directly down
As we will show throughout the paper, the full-QSO phase                     to the highest resolution level; while the former minimizes the
reaches a quasi-steady state over the whole hyper-refinement                 perturbations introduced, both methods yield similar long term
simulation time while accretion in the pre-QSO and                           behavior. We have also experimented with different functional
late-AGN phases is highly time variable.                                     forms for χref (R); we find that χref ∝ R2 provides a convenient
                                                                             mapping between standard resolution at 1 kpc and χref min such that
                          2.3. Hyper-Lagrangian refinement                   we reach uniform resolution at .100 pc.
   Lagrangian hydrodynamic methods are naturally adaptive,                      Table 1 summarizes the main properties of our fiducial sim-
populating high density regions with a larger number of gas                  ulations. The initial conditions are taken at each of the red-
resolution elements. In order to achieve super-Lagrangian re-                shift snapshots corresponding to the pre-QSO, full-QSO, and
finement in MFM, we split gas resolution elements progressively              late-AGN conditions, where we remove all black hole parti-
as they approach the central black hole. The distance R from                 cles other than the one located at the center of galaxy A4. The
each gas particle to the black hole is computed while traversing             actual black hole mass is set to MBH = 108 M such that we
Cosmological hyper-refinement simulations of quasar fueling                                                   7

          Table 1. Simulation parameters (units are physical): (1) Name: simulation designation. (2) z: initial redshift of hyper-refinement.
          (3) Mvir : halo virial mass. (4) M?: galaxy stellar mass. (5) Mgas : galaxy gas mass. (6) SFR: galaxy star formation rate. (7)
          ∆t: total duration (physical run-time) of the simulation at the full hyper-refinement level. (8) dtmin : minimum timestep of black
          hole particle. (9) mg : minimum mass of gas resolution element. (10) hsml min : minimum gas smoothing length.

           Name           z     Mvir [M ]     M? [M ]      Mgas [M ]     SFR [M yr−1 ]     ∆t [Myr]     dtmin [yr]   mg [M ]      min [pc]
                                                                                                                                 hsml
           pre-QSO       2.28    2.31e12       5.90e10      2.15e10            523             3          0.15          15         0.08
           full-QSO      2.25    2.37e12       8.21e10      1.87e10            222             4          0.42          30         0.08
           late-AGN      1.10    4.44e12       2.31e11       9.63e9             13             20         2.55          15         0.15

can compare the inflow rate under different conditions for the                  3.1. From zoom-in initial conditions to hyper-refined galactic
same MBH (we test different MBH values in additional simula-                                                 nuclei
tions; see §4.5). We use the same gravitational softenings for                    Figure 3 shows the gas distribution in the central 100 pc for the
dark matter, DM = 57 pc, but use fully-adaptive softenings for                three initial conditions considered, comparing the original reso-
gas (matching the hydrodynamic solver) with an arbitrarily small               lution from the parent cosmological zoom-in simulation (left) and
minimum allowed (but in practice minimum softenings reached                    that obtained in the cosmological hyper-refinement simulations
are given in Table 1), and treat the black hole correctly as a                 (right). Each pair of simulations is shown at the same evolu-
Keplerian point-like particle (using a much smaller softening,                 tion time, corresponding to ∼1 Myr (pre-QSO and full-QSO)
BH = 0.001 pc, intentionally smaller than the “accretion ra-                  and ∼6 Myr (late-AGN) after reaching the maximum resolu-
dius”). We adopt ? = Racc = 0.1 pc for stars, motivated roughly               tion in the hyper-refinement simulation. The top panels cor-
by the softening of gas at the densities where stars form in the               respond to the pre-QSO conditions at z = 2.28, with average
densest hyper-refined regions, but in Appendix A.2 we show that                Σgas ∼ 1010 M kpc−2 in the inner 100 pc. The initial conditions
our results are robust relative to changes in ?, including adap-              (top left) already capture a remarkable level of sub-structure for
tive and variable softenings. Hyper-refinement operates within                 a massive galaxy simulated in a full cosmological context, with
Rref = 1 kpc in all simulations presented here and we use fiducial             a clumpy, dense, circumnuclear ring and a hint of filamentary
maximum refinement factors χref  min = 1/1024 (full-QSO phase)
                                                                               structures that appear to connect with a central gas concentra-
and χrefmin = 1/2048 (pre-QSO and late-AGN phases). Addi-
                                                                               tion. The hyper-refinement simulation (top right) confirms the
tional simulations with lower refinement levels are employed for               presence of the ring-like structure in the nuclear region, which
resolution convergence tests (A.1). We save data snapshots for                 is now resolved by Ngas ∼ 2 × 106 resolution elements into a
every 0.01 Myr of evolution in all simulations. Except for the                 complex distribution of clumps, filaments, and spiral structures.
radical increase in resolution and the treatment of black hole                    The middle panels of Figure 3 show the nuclear gas distribu-
growth, our cosmological hyper-refinement simulations imple-                   tion for the full-QSO conditions at z = 2.25, characterized by an
ment identical physics as the original FIRE-2 simulation.                      extreme gas concentration reaching Σgas > 1011 M kpc−2 in the
    We reach minimum timesteps ∼0.1 yr and evolve our hyper-                   central 10 pc. The parent zoom-in simulation suggests the pres-
refinement simulations for ∆t ∼ 3–20 Myr (Table 1). The ini-                   ence of up to four nuclear spiral arms while the ×1000 increase in
tial conditions considered have characteristic dynamical times                 mass resolution yields a variety of morphological features with
tdyn ≡ R/vc ∼ 2 Myr → 0.2 Myr given the circular velocity                      more than three orders of magnitude in density contrast, includ-
vc on R ∼ 1 kpc → 100 pc scales. We are thus evolving the                      ing an ultra-dense eccentric disc within ∼ 10 pc that breaks up
simulations for ∼1–10 dynamical times at the scale where hyper-                into multiple spiral arms. Finally, the bottom panels correspond
refinement starts (1 kpc) and ∼10–100 dynamical times in the                   to the late-AGN conditions at z = 1.1, where the parent zoom-
nuclear region within which the maximum resolution is achieved                 in simulation shows a nearly homogeneous gas distribution with
(.100 pc). Our simulations are thus reaching a reasonable sta-                 average Σgas . 109 M kpc−2 in the inner 100 pc. The cosmolog-
tistical steady state in which the high resolution region matches              ical hyper-refinement simulation yields a remarkably different
the lower resolution conditions on larger scales in a physically               gas distribution, with a collection of dense clumps and filaments
meaningful way. In addition, we find characteristic flow times                 embedded in a low density medium and the formation of a central
tflow ≡ Mgas / MÛ in ∼ 6–22 Myr given the enclosed gas mass and                cavity extending a few tens of pc.
inflow rate at 1 kpc (§3.3), roughly similar to our total evolution
times, which emphasizes the importance of including larger scale
gas flows when evolving the nuclear region over many dynamical                              3.2. Mapping gas from Mpc to pc scales
times.                                                                           Figure 4 illustrates the full dynamic range of our simulations
                                                                               by showing the gas distribution for the pre-QSO conditions on
                                                                               different scales, from 1 Mpc to the central 10 pc. The top left
                                                                               panel shows the full extent of the zoom-in region, which is pop-
                  3. Overview of simulations                                   ulated with high resolution dark matter and gas elements and
8                                                       Daniel Anglés-Alcázar et al.

          mg = 3 × 104 M                    mg = 30 M                      On 10 kpc scales, the ISM shows a very irregular morphology,
                                                                        with gas surface densities in the range Σgas ∼ 107−10 M kpc−2 .
                                                                        The highest densities are reached within the inner 1 kpc in a
                                                                        clumpy ring-like structure of radius ∼ 400 pc, with a secondary
                                                                        gas concentration forming in the inner 100 pc. The black hole
                                                                        responds dynamically to the turbulent clumpy gas distribution
                                                                        by orbiting up to 10 pc away from the center of mass of the
                                                                        stellar component. Gravitational capture of infalling clumps and
                                                                        filaments yields pc-scale gas disks around the black hole with
    pre-QSO: z = 2.28                                                   changing orientation depending on the angular momentum of
                                                                        the infalling material.
               50 pc                           50 pc                       Figure 5 shows projected mass-weighted temperature distribu-
                                                                        tions from 1 Mpc to 10 pc scales for the full-QSO conditions at
                                                                        z = 2.25. In this case, the Mpc scale view in the top left panel
                                                                        highlights the thin, large scale filaments of cool gas (T < 105 K)
                                                                        feeding the central halo and the expanding hot gas (T > 106 K)
                                                                        heated by accretion shocks and galactic winds, reaching well
                                                                        beyond Rvir & 130 kpc. On 100 kpc scales, the CGM shows a
                                                                        prominent multi-phase structure with cold clumps and filaments
                                                                        embedded in a hot medium of virialized gas (T > 106.5 K). The
    full-QSO: z = 2.25                                                  projected temperature distribution in the central 10 kpc is highly
                                                                        irregular, with a large covering fraction of cold, star forming
                                                                        gas. Dust continuum radiative transfer calculations show that
                                                                        these conditions produce a submm-bright phase (Cochrane et al.
                                                                        2019). Zooming into the inner 1 kpc, cold gas with low angu-
                                                                        lar momentum falls toward the inner region while nuclear spiral
                                                                        arms form out of the turbulent ISM. In the central 100 pc, the sim-
                                                                        ulation captures ∼ 4 orders of magnitude variation in projected
                                                                        mass-weighted temperature, with prominent spiral structures of
                                                                        T ∼ 104−5 K gas embedded in T > 107 K gas heated by SNe.
    late-AGN: z = 1.10                                                  The bottom left panel highlights the temperature contrast in the
                                                                        inner 10 pc, where a cold (T ∼ 1000 K), pc-scale, rotationally
Figure 3. Projected gas mass surface density for a cubic volume of      supported gas disk forms around the black hole.
side 200 pc around the central black hole in the pre-QSO phase (top),      Figure 6 shows again projected gas surface density distribu-
full-QSO phase (middle), and late-AGN (bottom) simulations. Left        tions on multiple scales, as in Figure 4, but in this case for
panels correspond to the original FIRE-2 simulation chosen for re-      the late-AGN conditions at z = 1.1. The main halo has dou-
simulation, with mass resolution mb = 3×104 M (Anglés-Alcázar et al.    bled in mass since z ∼ 2.3, with Mvir = 4.4 × 1012 M and
2017c). Right panels show the gas distributions obtained at the same    Rvir ∼ 250 kpc, and the central galaxy has developed a massive
physical times in the hyper-refinement simulations with mb ∼ 30 M .     stellar component (M? = 2.3 × 1011 M ). The CGM on 100 kpc
The color scale is logarithmic from Σgas = 107.5 M kpc−2 (purple) to    scales shows a smoother gas distribution compared to earlier
1010.5 M kpc−2 (yellow).                                                times, though some dense gas structures are still prominent.
                                                                        With the gas fraction within 0.1 Rvir falling below fgas < 5% and
embedded in a low-resolution volume of [100 h−1 Mpc]3 that al-          SF R ∼ 10 M yr−1 , the galaxy forms a thin, rotationally sup-
lows us to capture tidal torques from large scale structures. High      ported disk in the inner ∼ 2 kpc. A low-density contrast, bipolar
density peaks correspond to the location of galaxies, tracing           structure appears to extend a few kpc above and below the plane
large scale filaments. The white dashed circle in the center in-        of the disk, likely driven by an earlier episode of galactic winds.
dicates the virial radius of the main halo, with Rvir = 130 kpc         Zooming into the central 1 kpc, the disk structure is dominated
and Mvir = 2.3 × 1012 M at z = 2.28. Zooming into the central           by star-forming regions with Σgas > 109 M kpc−2 embedded in
100 kpc, the CGM shows a complex gas distribution owing to              a low density medium, with a compact spiral structure superim-
the interplay between filamentary accretion from the IGM, cool-         posed. In the central 100 pc, gas consumption by star formation
ing of hot halo gas, and galactic winds from the central galaxy         and ejection by stellar feedback form a cavity that extends a few
and satellites (e.g. Hafen et al. 2019, 2020; Fielding et al. 2020).    tens of pc and starves the black hole of fuel (see also Anglés-
A merging satellite galaxy (M? ∼ 1010 M ) is approaching the            Alcázar et al. 2017c). Gravitational torques from the stellar com-
central (M? = 5.9 × 1010 M ), currently on its second passage           ponent drive gas clumps and filaments from the inner boundary
at a distance of ∼ 20 kpc and playing a minor role in driving           of the circumnuclear disk down to the central 10 pc, occasionally
instantaneous black hole growth (see §5.4).                             feeding the black hole (see §4.4).
Cosmological hyper-refinement simulations of quasar fueling                                                           9

  log10(    gas/M   kpc 2)                     1 Mpc                                                                     100 kpc                     9
                                                                                                                                                     8
      8                                                                                                                                              7
      7                                                                                                                                              6
      6
      5

                                                                                                                          10 kpc                     10
                                                                                                                                                     9
                                                                                                                                                     8
                                                                                                                                                     7

  z = 2.28
                      10 pc                                             100 pc                                             1 kpc

      11                                                                                          11                                                 10
      10                                                                                          10                                                 9
      9                                                                                           9                                                  8
                                                                                                  8

Figure 4. Multi-scale gas mass surface density distribution in a hyper-refinement simulation for pre-QSO conditions at z = 2.28. The top left panel
shows the central 1 Mpc, with the white dashed line indicating Rvir ∼ 130 kpc of the central halo. Subsequent panels progressively zoom into the central
10 pc of the main galaxy, with dynamic hyper-refinement occuring at  6 orders of magnitude variation in gas surface density (Σgas ∼ 105−11 M kpc−2 ) and a dynamic range spanning over 6 orders of magnitude
in spatial scale.
10                                                        Daniel Anglés-Alcázar et al.

  log10(T/k)                                  1 Mpc                                                                    100 kpc                     7
                                                                                                                                                   6
      7                                                                                                                                            5
      6                                                                                                                                            4
      5
      4

                                                                                                                        10 kpc                     7
                                                                                                                                                   6
                                                                                                                                                   5
                                                                                                                                                   4

  z = 2.25
                      10 pc                                            100 pc                                            1 kpc

      6                                                                                          7                                                 7
      5                                                                                          6                                                 6
      4                                                                                          5                                                 5
      3                                                                                          4                                                 4

Figure 5. Multi-scale mass-weighted temperature distribution for the full-QSO conditions at z = 2.25. The top left panel shows the central 1 Mpc,
with the white dashed line indicating the virial radius of the central halo. Subsequent panels progressively zoom into the central 10 pc, where the
location of the massive black hole is indicated by a + symbol. We model simultaneously large scale filaments of cool gas penetrating the virial shock of
dark matter halos all the way down to multi-phase ISM gas in the central 10 pc, where the black hole accretes from a cold, pc-scale gas disk embedded
in a hot medium.
Cosmological hyper-refinement simulations of quasar fueling                                                          11

  log10(    gas/M   kpc 2)                     1 Mpc                                                                     100 kpc                     8
                                                                                                                                                     7
      7                                                                                                                                              6
      6
      5

                                                                                                                          10 kpc                     9
                                                                                                                                                     8
                                                                                                                                                     7
                                                                                                                                                     6

  z = 1.10
                      10 pc                                             100 pc                                             1 kpc

      10                                                                                          10                                                 10
      9                                                                                           9                                                  9
      8                                                                                           8                                                  8
                                                                                                                                                     7

Figure 6. Multi-scale gas mass surface density distribution in the fiducial hyper-refinement simulation for the late-AGN conditions at z = 1.1. The
top left panel shows the central 1 Mpc with the white dashed line indicating Rvir ∼ 250 kpc of the central halo. Subsequent panels progressively zoom
into the central 10 pc of the main galaxy. The location of the central black hole is indicated by a + symbol in the lower left panel. A low density cavity
forms in the central 100 pc of the thin, rotationally supported gas disk, suppressing black hole growth. Gravitational instabilities in the inner edge of
the circumnuclear disk drive gas clumps and filaments toward the central 10 pc, occasionally feeding the black hole.
12                                                    Daniel Anglés-Alcázar et al.

          3.3. Galaxy radial structure and kinematics                                      15                       pre-QSO: z = 2.28, t = 3Myr
   Figure 7 shows the stellar, gas, and star formation rate surface                                                 full-QSO: z = 2.25, t = 4Myr
                                                                                           14                       late-AGN: z = 1.10, t = 20Myr

                                                                      log10( * /M kpc 2)
densities as a function of cylindrical radial distance Rcyl for the                        13                       before hyper-refinement
three conditions simulated. We define a cylindrical coordinate
system (centered on the black hole) independently for each time                            12
and radial bin, where the z-axis is aligned with the total angular                         11
momentum inside of the 3D radial distance R = Rcyl to the black                            10
hole, also performed independently for the gas and stellar com-                             9
ponents. Surface densities Σ?, Σgas , and ΣSFR are then computed
within these local cylindrical radial bins (logarithmically spaced)
                                                                                            8
by integrating the mass/SFR within ±100 pc along the vertical                              11
direction relative to the local cylindrical plane. This allows us
                                                                                           10

                                                                      kpc 2)
to capture spatial and temporal variations in the structure and
kinematics of the galaxy, but our results are not sensitive to the                         9

                                                                      gas/M
exact choice of local radial bins, their vertical extension rela-
tive to the local galaxy plane, or the use of spherical rather than                        8

                                                                      log10(
cylindrical radial bins. Note that the black hole and therefore the
adopted reference frame can move relative to the center of mass
                                                                                           7
of the galaxy (Figure 4), with subsequent analysis reflecting the                          6
instantaneous conditions affecting black hole growth.
   The pre-QSO conditions at z = 2.28 (green) show a sig-                                  7

                                                                           yr 1kpc 2)
nificant increase in stellar mass surface density Σ? ∼ 108 →                               6
1011 M kpc−2 in the range Rcyl = 1 → 0.1 kpc, with a mild                                  5
increase to Σ? ∼ 1011.7 M kpc−2 at smaller scales. The initial                             4
stellar distribution (dashed line) is very similar to the median Σ?
                                                                                           3
                                                                           SFR/M

found during the ∼ 3 Myr time evolution in the hyper-refinement
simulation (solid line), but it does not extend to within the in-                          2
                                                                           log10(

ner 1 pc. New stars formed out of ultra-high resolution gas                                1
populate the very inner region over the ∼ 3 Myr period. The                                0
shaded region indicates the 10–90% range for each radial bin,                                   4   3     2            1           0                1
reflecting the build up of stellar mass within 1 pc, reaching Σ? ∼                                      log10(Rcyl /kpc)
1012.5 M kpc−2 , but also variability in Σ? owing to black hole dy-
namics relative to the stellar distribution. The full-QSO phase        Figure 7. Stellar mass surface density (top), gas mass surface density
at z = 2.25 (orange), ∼40 Myr later, shows very similar Σ? at          (middle), and SFR surface density (bottom) as a function of cylindrical
> 100 pc but significantly higher values in smaller scales relative    distance Rcyl from the black hole (defined with respect to the angular mo-
to the pre-QSO conditions. The initial full-QSO conditions had         mentum axis of gas/stars at R < Rcyl ; see text) for the pre-QSO (green),
already reached Σ? ∼ 1013 M kpc−2 within ∼10 pc, increasing            full-QSO (orange), and late-AGN (blue) conditions. Solid lines show
to even more extreme densities as new ultra-high resolution stars      median values in each radial bin considering the full evolution time ∆t
form. Moving forward to z = 1.1, the late-AGN conditions               at hyper-refined resolution, while shaded regions indicate the 10–90%
(blue) show a more extended stellar distribution owing to size         percentile range reached in each radial bin. Dashed lines in the top
growth and expansion of the inner region. Within ∼100 pc, Σ?           panel correspond to the initial Σ? at the start of the hyper-refinement
is roughly ×10 lower compared to the full-QSO conditions but           simulations.
the original stellar distribution retained the central core with
Σ? ∼ 1013 M kpc−2 , further increasing during the ∼20 Myr              lower Σgas on galactic scales but more steady and significantly
hyper-refinement simulation period. The extreme stellar den-           larger gas and SFR surface densities (∼0.5 dex) at
Cosmological hyper-refinement simulations of quasar fueling                                       13

                                                                                                                                            T & 107 K consistently in the central ∼10–100 pc, where there
                                                                          8                                                                 is a stark contrast between the temperatures reached by the cold
   log10(T/K)

                                                                                                                                            and hot ISM phases (see Figure 5). Despite this, gas within
                                                                          6                                                                 ∼1 pc of the black hole is predominantly cold. By the time we
                                                                                                                                            reach the late-AGN conditions at z = 1.1, the galaxy has al-
                                                                          4                                                                 ready developed a hot halo and we find T ∼ 106.5 K at ∼10 kpc
                                                                                                                                            scales, dropping to T ∼ 104 K within the ∼ kpc scale gas disk. At
                                                                          3
 log10(v /km s 1) log10( r /km s 1) log10( z /km s 1) log10(cs /km s 1)

                                                                                                                                             1000 km s−1 with
                                                                          1                                                                 vφ ∝ R−1/2 , as expected for Keplerian rotation within the
                                                                                                                                            black hole radius of influence. With increased stellar and gas
                                                                          3                                                                 mass surface densities, the full-QSO conditions show even
                                                                                                                                            higher levels of turbulence. In this case, both σz and σr in-
                                                                          2                                                                 crease from ∼ 200 km s−1 on galaxy scales to ∼ 500 km s−1 at
                                                                                  pre-QSO: z = 2.28, t = 3Myr                               ∼ 0.1 pc. Nonetheless, rotational motion still dominates with
                                                                          1       full-QSO: z = 2.25, t = 4Myr                              vφ > 600 km s−1 at Rcyl < 100 pc. At later times, the rotational
                                                                                  late-AGN: z = 1.10, t = 20Myr
                                                                                                                                            velocity in the late-AGN conditions reaches vφ > 800 km s−1 on
                                                                              4           3             2            1     0        1       kpc scales owing to the build up of a massive stellar component.
                                                                                                      log10(Rcyl /kpc)
                                                                                                                                            The gas distribution shows a rotationally supported disk with
Figure 8. From top to bottom: average (mass-weighted) gas temper-                                                                           ∼ 50 km s−1 turbulent motions. The cool gas that penetrates the
ature, sound speed, vertical velocity dispersion, radial velocity disper-                                                                   inner hot cavity (blue shaded region within ∼10 pc) follows a
sion, and azimuthal velocity as a function of cylindrical radial distance                                                                   roughly Keplerian rotation curve with vφ ∝ R−1/2 .
from the black hole for the pre-QSO (green), full-QSO (orange), and                                                                            Overall, despite the high level of turbulent motions in the
late-AGN (blue) conditions. Solid lines show median values in each                                                                          pre-QSO and full-QSO conditions and the role of thermal
radial bin over the full evolution time ∆t, while shaded regions indicate                                                                   pressure in the late-AGN central cavity, rotational support ap-
the 10–90% percentile range reached in each radial bin.                                                                                     pears to dominate in the full range of scales analyzed, with
                                                                                                                                            cs < σz ∼ σr < vφ . As with the extreme stellar densities reached
                                                                                                                                            in the full-QSO conditions (Figure 7), higher than the maxi-
two panels show the average (mass-weighted) temperature (T)                                                                                 mum stellar surface density predicted by analytic models due to
and sound speed (cs ) within each cylindrical radial bin, extend-                                                                           the failure of stellar feedback (Hopkins et al. 2010; Grudić et al.
ing only ±10 pc along the vertical direction relative to the local                                                                          2019), the large vφ values at late times (higher than observed)
plane. As in Figure 7, solid lines show median values for the full                                                                          indicate the need for AGN feedback or some other additional
hyper-refinement simulation time and shaded regions indicate the                                                                            mechanism to suppress nuclear star formation (Wellons et al.
10–90% range variation in each radial bin. The pre-QSO con-                                                                                 2020; Parsotan et al. in prep.). We investigate next if large
ditions are dominated by cool gas (T . 105 K) in the entire                                                                                 inflow rates down to sub-pc scales can occur, a prerequisite for
range Rcyl = 0.1 pc–10 kpc owing to the large amount of dense                                                                               efficient “QSO-mode” black hole feedback.
gas and thus rapid cooling times. With average cs ∼ 10 km s−1 ,
thermal pressure support does not play a relevant dynamical
                                                                                                                                                               4. Multi-scale gas Inflows
role. Nonetheless, heating by SNe following intense star for-
mation can sometimes increase the temperature to T > 106 K                                                                                                 4.1. Explicit accretion at < 0.1 pc
in the inner ∼ 100 pc; note that the black hole can also move                                                                                 Figure 9 shows MÛ BH as a function of time measured at 0.1 pc in
from cold to hot dominated regions as it orbits within the central                                                                          hyper-refinement simulations for the pre-QSO (z = 2.28; green),
10 pc. The thermal properties in the full-QSO conditions are                                                                                full-QSO (z = 2.25; orange), and late-AGN (z = 1.10; blue)
rather different, with average T ∼ 106 K on galaxy scales and                                                                               conditions, considering in all cases a black hole with initial mass
14                                                                                    Daniel Anglés-Alcázar et al.
1.0
                            1 MBH = 0.59 M yr                                               pre-QSO
        log10(MBH/M yr 1)
                                                                        1

                                                                                                                 log10(MBH/M yr 1)
                                                                                                                                         1
                            0                                                                                                            0
                            1                                                                                                            1
0.8                         2                                                                                                            2   MBH = 5.82 M yr 1     full-QSO
                            0.0                 0.5               1.0       1.5   2.0          2.5     3.0                               0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
                                                                        t (Myr)                                                                           t (Myr)
                                                                                                                        z
                                10                 7 6             5        4           3                                            2                                                  1
                            1                        MBH = Mgas/tdyn
0.6
                            0                        MEdd
        log10(MBH/M yr 1)

                            1
                            2
                            3
                            4
0.4
                            5
                                                             1                    2                          3                                     4                   5                6
                                                                                                     Cosmictime (Gyr)

                                                                  MBH = 0.012 M yr
                                     log10(MBH/M yr 1)

                                                                                        1
                                                         1                                                                                                                   late-AGN
0.2
                                                         0
                                                         1
                                                         2
                                                         0.0                2.5         5.0           7.5                            10.0          12.5   15.0             17.5    20.0
0.0
                                                                                                                                     t (Myr)
   0.0                                                           0.2                          0.4                                            0.6                 0.8                        1.0
      Figure 9. Black hole accretion history from z = 10 → 1 inferred from sub-grid modeling in a full cosmological simulation (middle panel) compared
      to the explicit gas inflow rates measured at Racc = 0.1 pc in re-simulations with dynamic hyper-refinement (colored panels). The black solid line in
      the middle panel shows the free-fall accretion estimator MÛ BH = α Mgas /tdyn evaluated for the central black hole in halo A4 from Anglés-Alcázar
      et al. (2017c), where Mgas is the gas mass within R0 ∼ 100 pc, tdyn = (R03 /GMtot )1/2 , Mtot is the total mass within R0 , and α ≡ 5 × 10−4 is
      set such that MBH /Mbulge ∼ 0.002 at z = 1. The dashed gray line shows the Eddington rate ( MÛ Edd ) corresponding to the growing black hole.
      Vertical lines indicate the redshifts chosen for re-simulation with dynamic hyper-refinement for the pre-QSO (z = 2.28; green), full-QSO (z = 2.25;
      orange), and late-AGN (z = 1.10; blue) conditions. Inset panels show the explicit accretion histories measured at Racc = 0.1 pc in each case for
      re-simulations considering an initial black hole mass MBH  ini = 108 M , with horizontal gray dashed lines indicating M Û Edd (M ini ). Average accretion
                                                                                                                                      BH
      rates for re-simulation periods are shown in each inset panel and indicated as + symbols in the middle panel. Dynamic hyper-refinement simulations
      capture over four orders of magnitude variability in MÛ BH and predict qualitatively distinct phases of black hole growth, with (1) strong variability in
      pre-QSO conditions with MÛ BH = 0.001–10 M yr−1 , (2) steady super-Eddington feeding in the full-QSO phase (in the absence of AGN feedback),
      and (3) low luminosity in the late-AGN conditions with ∼25% duty cycle.
Cosmological hyper-refinement simulations of quasar fueling                                           15

                   3 pre-QSO: z = 2.28, t = 3Myr                               Figure 9 (green) shows high level of variability in accretion when
                                                                               measured at 0.1 pc, with MÛ BH = 0.001–10 M yr−1 and average
                   2                                                                                                                              ini =
                                                                               h MÛ BH i = 0.6 M yr−1 during a period of ∼3 Myr. For MBH
 log10(M/M yr 1)

                   1                                                               8
                                                                               10 M , this implies an average Eddington ratio hλEdd i ≈ 0.3,
                   0                                                           where λEdd ≡ MÛ BH / MÛ Edd and MÛ Edd ≈ 2.2 M yr−1 is the Edding-
                   1                                                           ton rate. The full-QSO phase at z = 2.25 (orange panel; top
                                                                               right) coincides with the peak of black hole accretion according
                   2
                                                                               to the free-fall sub-grid accretion model, occurring ∼40 Myr after
                   3                                                           the pre-QSO conditions. In this case, the hyper-refinement sim-
                   3 full-QSO: z = 2.25, t = 4Myr                              ulation predicts h MÛ BH i ≈ 6 M yr−1 with very little variability
                   2                                                           during a period of ∼4 Myr, corresponding to hλEdd i ≈ 2.6. This
                                                                               implies that quasi-steady super-Eddington feeding rates onto the
 log10(M/M yr 1)

                   1                                                           accretion disk are possible in the absence of AGN feedback. The
                   0                                                           late-AGN conditions at z = 1.1 (blue panel; bottom) show sig-
                   1                                                           nificantly lower accretion rate, with h MÛ BH i = 0.01 M yr−1 dur-
                                                          Enclosed SFR         ing a period of ∼20 Myr, corresponding to hλEdd i ≈ 0.005. The
                   2                                      Inflow rate          formation of a low-density central cavity filled with T ∼ 107 K
                   3                                      Outflow rate         gas (Figure 8) inhibits black hole growth during ∼75% of the time
                   2 late-AGN: z = 1.10, t = 20Myr                             (with MÛ BH = 0), while the instantaneous accretion rate ranges
                                                                               from MÛ BH = 0.001–0.1 M yr−1 when cold gas clumps and fila-
                   0                                                           ments destabilize from the inner edge of the circumnuclear disk
 log10(M/M yr 1)

                                                                               and fall toward the black hole (Figure 6). The finite mass of
                                                                               gas resolution elements (mg ∼ 15 M ) implies an upper limit of
                   2
                                                                               roughly MÛ BH < 10−6 M yr−1 during the inactive periods in the
                                                                               late-AGN phase.
                   4                                                              Overall, our hyper-refinement simulations predict qualitatively
                                                                               distinct accretion phases with properties that are not trivial to
                       4        3          2          1     0            1     infer even from the ∼100 pc physical conditions available in state-
                                          log10(R/kpc)                         of-the-art cosmological zoom-in simulations. Remarkably, we
Figure 10. Time-averaged gas mass flow rate as a function of spherical         show that it is possible to feed the central 0.1 pc at a rate sufficient
radial distance R from the central black hole for the pre-QSO (top),
                                                                               to power a luminous QSO (Lbol ≈ 3 × 1046 erg s−1 for MÛ BH =
                                                                               6 M yr−1 and rad = 0.1) during the epoch of peak nuclear gas
full-QSO (middle), and late-AGN (bottom) conditions. Black solid
                                                                               density in the central galaxy of a massive halo (Mvir ∼ 1012.5 M )
and dotted lines indicate radial bins with net inflow and outflow rate
                                                                               at z ∼ 2. At the same time, our simulations predict strong
respectively. In each panel, the colored dashed line shows the time-
                                                                               variability in the conditions preceding the QSO phase and low-
averaged enclosed SFR and the shaded region indicates the 10–90%
                                                                               duty cycle and lower luminosity for the more common conditions
percentile range achieved in each radial bin during the full evolution         prevalent at later times.
time ∆t. The enclosed SFR generally traces the radial flow rate profile,
with some indication for comparatively lower star formation at < 1 pc.                       4.2. Star formation–inflow connection
                                                                                  Figure 10 investigates the connection between gas inflow rate
MBHini = 108 M . For comparison, the middle panel shows the                    and SFR over five orders of magnitude in spatial scales. For each
black hole accretion history from z = 10 → 1 obtained in the                   hyper-refinement simulation, we compute the net mass flow rate
original cosmological zoom-in simulation by adopting the “free-                across spherical shells and show the time-averaged radial profile
fall" sub-grid accretion estimator MÛ BH = α Mgas /tdyn , where                as either solid or dotted lines for net inflow and outflow rate,
tdyn = (R03 /GMtot )1/2 is the dynamical time within R0 ∼ 100 pc               respectively. Note that large-scale, coherent galactic outflow
and Mgas and Mtot are the gas mass and total mass within R0 .                  events driven by stellar feedback are common at higher redshift
Here, we define α ≡ 5 × 10−4 such that MBH /Mbulge ∼ 0.002                     but absent in the conditions analyzed here, where inflows and
at z = 1, but the characteristic accretion history obtained is                 outflows can coexist in the same radial domain. Colored dashed
independent of this choice. In the absence of AGN feedback,                    lines show the time-averaged enclosed SFR as a function of radial
the sub-grid accretion model predicts short accretion episodes                 distance, computed from the instantaneous SFR of individual gas
that can reach or even exceed the Eddington rate at early times                elements, and shaded regions indicate the 10–90% variation in
(z > 2.3) and a transition to more steady accretion at late times,             each radial bin.
resembling the evolution in nuclear gas surface density shown in                  In the pre-QSO conditions (top), inflows dominate over out-
Figure 1 (see also Anglés-Alcázar et al. 2017c).                               flows in the entire radial range R = 0.1 pc–10 kpc, with the peak
   The pre-QSO conditions at z = 2.28 target the epoch when the                inflow rate reaching MÛ in ∼ 475 M yr−1 at ∼1 kpc and a mono-
bursty-to-steady transition is happening. The top left panel of                tonic decrease to MÛ in . 1 M yr−1 at ∼0.1 pc. The enclosed SFR
                                                                               shows significant variability occurring during only ∼3 Myr, in-
You can also read